Search in sources :

Example 91 with HoodieWriteConfig

use of org.apache.hudi.config.HoodieWriteConfig in project hudi by apache.

the class TestHoodieBulkInsertDataInternalWriter method testGlobalFailure.

/**
 * Issue some corrupted or wrong schematized InternalRow after few valid InternalRows so that global error is thrown. write batch 1 of valid records write batch2 of invalid records which is expected
 * to throw Global Error. Verify global error is set appropriately and only first batch of records are written to disk.
 */
@Test
public void testGlobalFailure() throws Exception {
    // init config and table
    HoodieWriteConfig cfg = getWriteConfig(true);
    HoodieTable table = HoodieSparkTable.create(cfg, context, metaClient);
    String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[0];
    String instantTime = "001";
    HoodieBulkInsertDataInternalWriter writer = new HoodieBulkInsertDataInternalWriter(table, cfg, instantTime, RANDOM.nextInt(100000), RANDOM.nextLong(), RANDOM.nextLong(), STRUCT_TYPE, true, false);
    int size = 10 + RANDOM.nextInt(100);
    int totalFailures = 5;
    // Generate first batch of valid rows
    Dataset<Row> inputRows = getRandomRows(sqlContext, size / 2, partitionPath, false);
    List<InternalRow> internalRows = toInternalRows(inputRows, ENCODER);
    // generate some failures rows
    for (int i = 0; i < totalFailures; i++) {
        internalRows.add(getInternalRowWithError(partitionPath));
    }
    // generate 2nd batch of valid rows
    Dataset<Row> inputRows2 = getRandomRows(sqlContext, size / 2, partitionPath, false);
    internalRows.addAll(toInternalRows(inputRows2, ENCODER));
    // issue writes
    try {
        for (InternalRow internalRow : internalRows) {
            writer.write(internalRow);
        }
        fail("Should have failed");
    } catch (Throwable e) {
    // expected
    }
    BaseWriterCommitMessage commitMetadata = (BaseWriterCommitMessage) writer.commit();
    Option<List<String>> fileAbsPaths = Option.of(new ArrayList<>());
    Option<List<String>> fileNames = Option.of(new ArrayList<>());
    // verify write statuses
    assertWriteStatuses(commitMetadata.getWriteStatuses(), 1, size / 2, false, fileAbsPaths, fileNames);
    // verify rows
    Dataset<Row> result = sqlContext.read().parquet(fileAbsPaths.get().toArray(new String[0]));
    assertOutput(inputRows, result, instantTime, fileNames, true);
}
Also used : HoodieWriteConfig(org.apache.hudi.config.HoodieWriteConfig) HoodieTable(org.apache.hudi.table.HoodieTable) ArrayList(java.util.ArrayList) List(java.util.List) InternalRow(org.apache.spark.sql.catalyst.InternalRow) Row(org.apache.spark.sql.Row) InternalRow(org.apache.spark.sql.catalyst.InternalRow) Test(org.junit.jupiter.api.Test) ParameterizedTest(org.junit.jupiter.params.ParameterizedTest)

Example 92 with HoodieWriteConfig

use of org.apache.hudi.config.HoodieWriteConfig in project hudi by apache.

the class TestHoodieDataSourceInternalWriter method testDataSourceWriterInternal.

private void testDataSourceWriterInternal(Map<String, String> extraMetadata, Map<String, String> expectedExtraMetadata, boolean populateMetaFields) throws Exception {
    // init config and table
    HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
    String instantTime = "001";
    // init writer
    HoodieDataSourceInternalWriter dataSourceInternalWriter = new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(extraMetadata), populateMetaFields, false);
    DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(0, RANDOM.nextLong(), RANDOM.nextLong());
    String[] partitionPaths = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS;
    List<String> partitionPathsAbs = new ArrayList<>();
    for (String partitionPath : partitionPaths) {
        partitionPathsAbs.add(basePath + "/" + partitionPath + "/*");
    }
    int size = 10 + RANDOM.nextInt(1000);
    int batches = 2;
    Dataset<Row> totalInputRows = null;
    for (int j = 0; j < batches; j++) {
        String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
        Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
        writeRows(inputRows, writer);
        if (totalInputRows == null) {
            totalInputRows = inputRows;
        } else {
            totalInputRows = totalInputRows.union(inputRows);
        }
    }
    HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
    List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
    commitMessages.add(commitMetadata);
    dataSourceInternalWriter.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
    metaClient.reloadActiveTimeline();
    Dataset<Row> result = HoodieClientTestUtils.read(jsc, basePath, sqlContext, metaClient.getFs(), partitionPathsAbs.toArray(new String[0]));
    // verify output
    assertOutput(totalInputRows, result, instantTime, Option.empty(), populateMetaFields);
    assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
    // verify extra metadata
    Option<HoodieCommitMetadata> commitMetadataOption = HoodieClientTestUtils.getCommitMetadataForLatestInstant(metaClient);
    assertTrue(commitMetadataOption.isPresent());
    Map<String, String> actualExtraMetadata = new HashMap<>();
    commitMetadataOption.get().getExtraMetadata().entrySet().stream().filter(entry -> !entry.getKey().equals(HoodieCommitMetadata.SCHEMA_KEY)).forEach(entry -> actualExtraMetadata.put(entry.getKey(), entry.getValue()));
    assertEquals(actualExtraMetadata, expectedExtraMetadata);
}
Also used : InternalRow(org.apache.spark.sql.catalyst.InternalRow) Arrays(java.util.Arrays) Dataset(org.apache.spark.sql.Dataset) HoodieTestDataGenerator(org.apache.hudi.common.testutils.HoodieTestDataGenerator) Option(org.apache.hudi.common.util.Option) HashMap(java.util.HashMap) Disabled(org.junit.jupiter.api.Disabled) DataSourceWriteOptions(org.apache.hudi.DataSourceWriteOptions) DataWriter(org.apache.spark.sql.sources.v2.writer.DataWriter) ArrayList(java.util.ArrayList) Map(java.util.Map) DataSourceOptions(org.apache.spark.sql.sources.v2.DataSourceOptions) Assertions.assertEquals(org.junit.jupiter.api.Assertions.assertEquals) MethodSource(org.junit.jupiter.params.provider.MethodSource) ENCODER(org.apache.hudi.testutils.SparkDatasetTestUtils.ENCODER) HoodieWriteConfig(org.apache.hudi.config.HoodieWriteConfig) HoodieCommitMetadata(org.apache.hudi.common.model.HoodieCommitMetadata) Row(org.apache.spark.sql.Row) STRUCT_TYPE(org.apache.hudi.testutils.SparkDatasetTestUtils.STRUCT_TYPE) Arguments(org.junit.jupiter.params.provider.Arguments) Test(org.junit.jupiter.api.Test) ParameterizedTest(org.junit.jupiter.params.ParameterizedTest) List(java.util.List) Stream(java.util.stream.Stream) SparkDatasetTestUtils.toInternalRows(org.apache.hudi.testutils.SparkDatasetTestUtils.toInternalRows) Assertions.assertTrue(org.junit.jupiter.api.Assertions.assertTrue) SparkDatasetTestUtils.getRandomRows(org.apache.hudi.testutils.SparkDatasetTestUtils.getRandomRows) HoodieClientTestUtils(org.apache.hudi.testutils.HoodieClientTestUtils) Collections(java.util.Collections) HashMap(java.util.HashMap) ArrayList(java.util.ArrayList) HoodieWriteConfig(org.apache.hudi.config.HoodieWriteConfig) HoodieCommitMetadata(org.apache.hudi.common.model.HoodieCommitMetadata) DataSourceOptions(org.apache.spark.sql.sources.v2.DataSourceOptions) InternalRow(org.apache.spark.sql.catalyst.InternalRow) Row(org.apache.spark.sql.Row) InternalRow(org.apache.spark.sql.catalyst.InternalRow)

Example 93 with HoodieWriteConfig

use of org.apache.hudi.config.HoodieWriteConfig in project hudi by apache.

the class TestHoodieDataSourceInternalWriter method testAbort.

/**
 * Tests that DataSourceWriter.abort() will abort the written records of interest write and commit batch1 write and abort batch2 Read of entire dataset should show only records from batch1.
 * commit batch1
 * abort batch2
 * verify only records from batch1 is available to read
 */
@ParameterizedTest
@MethodSource("bulkInsertTypeParams")
public void testAbort(boolean populateMetaFields) throws Exception {
    // init config and table
    HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
    String instantTime0 = "00" + 0;
    // init writer
    HoodieDataSourceInternalWriter dataSourceInternalWriter = new HoodieDataSourceInternalWriter(instantTime0, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
    DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(0, RANDOM.nextLong(), RANDOM.nextLong());
    List<String> partitionPaths = Arrays.asList(HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS);
    List<String> partitionPathsAbs = new ArrayList<>();
    for (String partitionPath : partitionPaths) {
        partitionPathsAbs.add(basePath + "/" + partitionPath + "/*");
    }
    int size = 10 + RANDOM.nextInt(100);
    int batches = 1;
    Dataset<Row> totalInputRows = null;
    for (int j = 0; j < batches; j++) {
        String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
        Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
        writeRows(inputRows, writer);
        if (totalInputRows == null) {
            totalInputRows = inputRows;
        } else {
            totalInputRows = totalInputRows.union(inputRows);
        }
    }
    HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
    List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
    commitMessages.add(commitMetadata);
    // commit 1st batch
    dataSourceInternalWriter.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
    metaClient.reloadActiveTimeline();
    Dataset<Row> result = HoodieClientTestUtils.read(jsc, basePath, sqlContext, metaClient.getFs(), partitionPathsAbs.toArray(new String[0]));
    // verify rows
    assertOutput(totalInputRows, result, instantTime0, Option.empty(), populateMetaFields);
    assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
    // 2nd batch. abort in the end
    String instantTime1 = "00" + 1;
    dataSourceInternalWriter = new HoodieDataSourceInternalWriter(instantTime1, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
    writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(1, RANDOM.nextLong(), RANDOM.nextLong());
    for (int j = 0; j < batches; j++) {
        String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
        Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
        writeRows(inputRows, writer);
    }
    commitMetadata = (HoodieWriterCommitMessage) writer.commit();
    commitMessages = new ArrayList<>();
    commitMessages.add(commitMetadata);
    // commit 1st batch
    dataSourceInternalWriter.abort(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
    metaClient.reloadActiveTimeline();
    result = HoodieClientTestUtils.read(jsc, basePath, sqlContext, metaClient.getFs(), partitionPathsAbs.toArray(new String[0]));
    // verify rows
    // only rows from first batch should be present
    assertOutput(totalInputRows, result, instantTime0, Option.empty(), populateMetaFields);
}
Also used : ArrayList(java.util.ArrayList) HoodieWriteConfig(org.apache.hudi.config.HoodieWriteConfig) DataSourceOptions(org.apache.spark.sql.sources.v2.DataSourceOptions) InternalRow(org.apache.spark.sql.catalyst.InternalRow) Row(org.apache.spark.sql.Row) InternalRow(org.apache.spark.sql.catalyst.InternalRow) ParameterizedTest(org.junit.jupiter.params.ParameterizedTest) MethodSource(org.junit.jupiter.params.provider.MethodSource)

Example 94 with HoodieWriteConfig

use of org.apache.hudi.config.HoodieWriteConfig in project hudi by apache.

the class TestHoodieDataSourceInternalWriter method testMultipleDataSourceWrites.

@ParameterizedTest
@MethodSource("bulkInsertTypeParams")
public void testMultipleDataSourceWrites(boolean populateMetaFields) throws Exception {
    // init config and table
    HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
    int partitionCounter = 0;
    // execute N rounds
    for (int i = 0; i < 2; i++) {
        String instantTime = "00" + i;
        // init writer
        HoodieDataSourceInternalWriter dataSourceInternalWriter = new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
        List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
        Dataset<Row> totalInputRows = null;
        DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(partitionCounter++, RANDOM.nextLong(), RANDOM.nextLong());
        int size = 10 + RANDOM.nextInt(1000);
        // one batch per partition
        int batches = 2;
        for (int j = 0; j < batches; j++) {
            String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
            Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
            writeRows(inputRows, writer);
            if (totalInputRows == null) {
                totalInputRows = inputRows;
            } else {
                totalInputRows = totalInputRows.union(inputRows);
            }
        }
        HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
        commitMessages.add(commitMetadata);
        dataSourceInternalWriter.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
        metaClient.reloadActiveTimeline();
        Dataset<Row> result = HoodieClientTestUtils.readCommit(basePath, sqlContext, metaClient.getCommitTimeline(), instantTime, populateMetaFields);
        // verify output
        assertOutput(totalInputRows, result, instantTime, Option.empty(), populateMetaFields);
        assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
    }
}
Also used : ArrayList(java.util.ArrayList) HoodieWriteConfig(org.apache.hudi.config.HoodieWriteConfig) DataSourceOptions(org.apache.spark.sql.sources.v2.DataSourceOptions) InternalRow(org.apache.spark.sql.catalyst.InternalRow) Row(org.apache.spark.sql.Row) InternalRow(org.apache.spark.sql.catalyst.InternalRow) ParameterizedTest(org.junit.jupiter.params.ParameterizedTest) MethodSource(org.junit.jupiter.params.provider.MethodSource)

Example 95 with HoodieWriteConfig

use of org.apache.hudi.config.HoodieWriteConfig in project hudi by apache.

the class TestHoodieDataSourceInternalWriter method testLargeWrites.

// takes up lot of running time with CI.
@Disabled
@ParameterizedTest
@MethodSource("bulkInsertTypeParams")
public void testLargeWrites(boolean populateMetaFields) throws Exception {
    // init config and table
    HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
    int partitionCounter = 0;
    // execute N rounds
    for (int i = 0; i < 3; i++) {
        String instantTime = "00" + i;
        // init writer
        HoodieDataSourceInternalWriter dataSourceInternalWriter = new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
        List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
        Dataset<Row> totalInputRows = null;
        DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(partitionCounter++, RANDOM.nextLong(), RANDOM.nextLong());
        int size = 10000 + RANDOM.nextInt(10000);
        // one batch per partition
        int batches = 3;
        for (int j = 0; j < batches; j++) {
            String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
            Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
            writeRows(inputRows, writer);
            if (totalInputRows == null) {
                totalInputRows = inputRows;
            } else {
                totalInputRows = totalInputRows.union(inputRows);
            }
        }
        HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
        commitMessages.add(commitMetadata);
        dataSourceInternalWriter.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
        metaClient.reloadActiveTimeline();
        Dataset<Row> result = HoodieClientTestUtils.readCommit(basePath, sqlContext, metaClient.getCommitTimeline(), instantTime, populateMetaFields);
        // verify output
        assertOutput(totalInputRows, result, instantTime, Option.empty(), populateMetaFields);
        assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
    }
}
Also used : ArrayList(java.util.ArrayList) HoodieWriteConfig(org.apache.hudi.config.HoodieWriteConfig) DataSourceOptions(org.apache.spark.sql.sources.v2.DataSourceOptions) InternalRow(org.apache.spark.sql.catalyst.InternalRow) Row(org.apache.spark.sql.Row) InternalRow(org.apache.spark.sql.catalyst.InternalRow) ParameterizedTest(org.junit.jupiter.params.ParameterizedTest) MethodSource(org.junit.jupiter.params.provider.MethodSource) Disabled(org.junit.jupiter.api.Disabled)

Aggregations

HoodieWriteConfig (org.apache.hudi.config.HoodieWriteConfig)327 Test (org.junit.jupiter.api.Test)179 ParameterizedTest (org.junit.jupiter.params.ParameterizedTest)173 HoodieRecord (org.apache.hudi.common.model.HoodieRecord)169 ArrayList (java.util.ArrayList)136 List (java.util.List)133 SparkRDDWriteClient (org.apache.hudi.client.SparkRDDWriteClient)126 HoodieTable (org.apache.hudi.table.HoodieTable)117 HoodieTableMetaClient (org.apache.hudi.common.table.HoodieTableMetaClient)111 HashMap (java.util.HashMap)93 Path (org.apache.hadoop.fs.Path)92 WriteStatus (org.apache.hudi.client.WriteStatus)86 HoodieInstant (org.apache.hudi.common.table.timeline.HoodieInstant)84 Collectors (java.util.stream.Collectors)81 Map (java.util.Map)76 HoodieTestDataGenerator (org.apache.hudi.common.testutils.HoodieTestDataGenerator)76 Assertions.assertEquals (org.junit.jupiter.api.Assertions.assertEquals)74 Arrays (java.util.Arrays)73 HoodieSparkTable (org.apache.hudi.table.HoodieSparkTable)72 Option (org.apache.hudi.common.util.Option)69